Wikipedia:Dispute resolution noticeboard/Archive 182

Archive 175Archive 180Archive 181Archive 182Archive 183Archive 184Archive 185

Natalia Toreeva

  – General close. See comments for reasoning.
Closed discussion

Akaike information criterion

  – New discussion.

Have you discussed this on a talk page?

Yes, I have discussed this issue on a talk page already.

Location of dispute

Users involved

Dispute overview

For a month now, there has been a content dispute between BetterMath and myself, about how to define a particular statistic called the Akaike information criterion (AIC) in a precise and intuitive way in the article's opening sentence. As of right now, the article describes the AIC as "an estimator of the relative quality of statistical models," which in my opinion is unnecessarily vague because "quality of a model" is so intangible that it is not even wrong. To give an analogy, it is a little bit like calling the Human Development Index (HDI) "an estimator of the relative quality of a country." Obviously the current lead of HDI describes it more precisely as a "composite index of life expectancy, education, and per capita income indicators," simply because this is what the HDI actually measures, i.e. the proxies of economic development. Similarly, the AIC approximates the "quality of a model" from prediction (or forecasting) error. The closer on target, the better the model. Because this is what AIC measures, I wanted to amend the current opening sentence with five words to "...an estimator of out-of-sample prediction error and thereby the relative quality...", but BetterMath objected.

My contribution is supported by two reliable sources that confirm the proposed definition nearly verbatim. BetterMath has brushed them aside, claiming them to be "undergraduate textbooks" (which in fact neither of them is) that contained "errors." Trying to point out those errors, BetterMath demonstrated a stunning lack of understanding of basic statistical concepts, like the difference between "in-sample" and "out-of-sample" statistics, which (after a month of good-faith discussion) makes me seriously doubt his ability to accurately judge the subject at hand. This content dispute is in dire need of a third opinion.

How have you tried to resolve this dispute before coming here?

Talk:Akaike_information_criterion#"relative_quality_of_statistical_models"

How do you think we can help resolve the dispute?

Ideally, somebody with sufficient knowledge in statistics could settle the issue by deciding whether the proposed definition is correct or not. But apart from that, generally any third person could look at this content dispute and decide on whose side the "burden of proof" now falls. In addition to the two sources referenced in the article, I have presented three more on the article talk page, whereas BetterMath has yet to produce a single source that contradicts any of the once listed.

Summary of dispute by BetterMath

For anyone involved in this dispute, it would be helpful to have some understanding of statistical models. An informal introduction is enough, and is given by Statistical model#Introduction. (Perhaps read that before reading the following.)

Assume that we have some data. Suppose that we also have several statistical models of the data. Some of those models will be better than others for modeling the data. We want to select the model that is in some sense the best model for the data. AIC provides a method for selecting. (It is based on information theory.)

Once we have selected a model, we can use the model to make predictions (also called forecasts). Predictions will rarely be perfect; rather, they will usually have some error.

Suppose that the data points are statistically independent. Then, it can be shown that AIC has a nice property: AIC selects the model that has the minimal expected error when making predictions.

The main issue on which Bender235 and I are in dispute is the following.

If the data points are not independent, then in general, we do not know what properties the prediction errors have. Bender235, however, does not accept that; instead, Bender235 claims that every model selected via AIC has the above-noted nice property. I have explained the issue in detail, and provided several reliable sources. All to only little effect.

In addition to the above issue, Bender235 has claimed that using the term quality of a model in the lead of the article is “unnecessarily vague”. The term is defined informally in the lead (second paragraph). A more formal definition is in the first section of the article: Akaike information criterion#Definition. That definition is somewhat technical (relying on the Kullback–Leibler divergence). I believe that the current approach, giving an informal definition in the lead, well adheres to WP:Lead#Provide an accessible overview.

BetterMath (talk) 20:17, 20 October 2019 (UTC)

Akaike information criterion discussion

Please keep discussion to a minimum before being opened by a volunteer. Continue on article talk page if necessary.

@BetterMath: I feel like this dispute can actually resolved very easily: in a situation with statistically dependent observations, imagine we have two competing models, and we calculate AIC for each of them. The one with lower AIC is the "better" one, but better in what sense? What does it do better than the other model? The answer is either (a) the "better" model predicts better, i.e. with less error, or (b) we don't know, because AIC is not applicable here, so whatever value it has is irrelevant and it cannot be used for model selection. Which one is it? --bender235 (talk) 20:47, 20 October 2019 (UTC)

Shahrbanu

  – New discussion.
Filed by LissanX on 23:07, 3 November 2019 .

Have you discussed this on a talk page?

Yes, though only one user, Kansas Bear (talk · contribs), has commented.

Location of dispute

Users involved

Dispute overview

Some users have been wiping content to enforce a one-sided non-neutral view. This line, complete with multiple sources, is consistently removed:

The date of reports, however, coincide with the golden age of Hadith compilations, with the Shia Four Books and Sunni Six Books all being released around the 9th century, as written Islamic reports were sparse prior to this period.

From this paragraph:

Islamic writers, such as al-Mubarrad, Ya'qubi and al-Kulayni, wrote the earliest reports of Shahrbanu and her imperial Persian background from the 9th century onward. However, the earliest sources make no mention of the mother of Ali ibn Husayn, nor do they ascribe him with maternal royal ancestry. The first references were from Ibn Sa'd and Ibn Qutaybah, also in the 9th century, who instead describe her as being a slave from Sindh. The Encyclopædia Iranica alleges that Shahrbanu was "undeniably legendary".

It was removed twice (1 and 2) for two completely different purported reasons. It’s clear that these revisions are simply to censor the fact that the 9th century Islamic texts are credible and are not necessarily unreliable as the mutilated version of the paragraph suggests.

The content of the added content is thoroughly corroborated by the relevant facts, such as the following dates of the Hadith compilations referenced:

The Six Books:

The Four Books:

And the fact that the earliest known Hadith compilations were Kitab Sulaym ibn Qays by Sulaym ibn Qays (died c. 689-709) for Shias and Sahifah Hammam ibn Munabbih by Hammam ibn Munabbih (died 719 CE) for Sunnis.

These facts were substantiated by comprehensive citations.

In addition to this, the translation of name of the person whom the article is about is also arbitrarily reverted each time.

Attempts to take personal control of the article has meant that they are not even willing to tolerate one line on the same paragraph, not even a simple name translation.

How have you tried to resolve this dispute before coming here?

Talk:Shahrbanu#Censoring_Content

User_talk:Alivardi#Arbitrarily_wiping_content_on_Shahrbanu

How do you think we can help resolve the dispute?

I would like for the content to be allowed for fair and neutral content to be displayed in the article. To have multiple claims of the unreliability of 9th century texts be presented, while removing one line saying that 9th century texts are also considered credible to provide a complete, appropriate outlook, is disruptive editing. — LissanX (talk) 23:07, 3 November 2019 (UTC)

  • Volunteer Note - The filing editor has not yet notified the other two editors of this filing. Please do not refer to a content dispute about the reliability of sources as censorship. Please read Yelling Censorship, and wait for a volunteer to facilitate a discussion of the reliability of sources. Robert McClenon (talk) 20:52, 4 November 2019 (UTC)

Stephen McAllister_(lawyer)

  – General close. See comments for reasoning.
Closed discussion

Jack Buckby

  – General close. See comments for reasoning.
Closed discussion

Call-out culture

  – Closed as failed. See comments for reasoning.
Closed discussion

Talk:Horn of_Africa#Jebel_Irhoud_in_Morocco_obsession

  – General close. See comments for reasoning.
Closed discussion

Goddard Gunster

  – General close. See comments for reasoning.
Closed discussion

Boyfriend loophole

  – General close. See comments for reasoning.
Closed discussion

Vedanta#History

  – General close. See comments for reasoning.
Closed discussion

Proclamation of Indonesian Independence

  – General close. See comments for reasoning.
Closed discussion

British Rail Class 390

  – Closed as failed. See comments for reasoning.
Closed discussion

Law enforcement_in_New_York

  – General close. See comments for reasoning.
Closed discussion

Cow vigilante violence in India

  – General close. See comments for reasoning.
Closed discussion

Cushitic peoples

  – General close. See comments for reasoning.
Closed discussion

Joe Sanchez

  – General close. See comments for reasoning.
Closed discussion

List of Steven Universe episodes

  – General close. See comments for reasoning.
Closed discussion

Opinion polling for the next Spanish general election

  Dispute resolved successfully. See comments for reasoning.
Closed discussion

Stephen McAllister_(lawyer)

  – General close. See comments for reasoning.
Closed discussion

Massimo Introvigne

  – General close. See comments for reasoning.
Closed discussion

2019 United Kingdom general election

  – General close. See comments for reasoning.
Closed discussion

Jo Swinson

  – General close. See comments for reasoning.
Closed discussion

White Helmets (Syrian Civil War)

  – General close. See comments for reasoning.
Closed discussion

Chaim Shacham

  – General close. See comments for reasoning.
Closed discussion

Feminazi

  – General close. See comments for reasoning.
Closed discussion